Dear David,

The 'constructional polysemy', from Goldberg's Constructional Grammar, is possible, though not as a 'transfer of an object between an agent and a receiver', as she believes. It does not necessarily have, on one hand, lexical items, and on the other, a syntax diagram. In reality it is a specific syntactic-semantic schema of a particular verb, which is assigned a schematic characterization of predicate arguments. A paradigmatic example of this is the ditransitive constructional schema (English double object or direct object and indirect object in Spanish); Now, here the meaning is not independent of lexical items, as Goldberg argues; and this for two reasons: first, because you can check the existence of a central or prototypical meaning, and second, because, as argued Croft, the meanings given by Goldberg are not strictly provided by the syntactic construction, but also by the verb . Then we can say that, in cases of ambiguity, as is the use of ditransitive verbs, we could accept, not a hierarchical distribution (vertical) between levels, but horizontally (heterarchical). Therefore the form and the meaning can be seen as complementary, and not as exclusive.

I hope this answer helps to answer your question.

It is important to distinguish between "Construction Grammar" proper (as developed by Fillmore & Kay) and "construction grammar" in the more general sense (i.e., the various construction grammar variants accepted as the model of grammar by most who adopt the cognitive linguistic framework; the more general term includes both explicit adoption of some construction grammar such as Croft's or Goldberg's but also the adoption of these, variants, and/or comparable models exemplified in works by Lakoff, Langacker, Bybee, Wierzbicka, Hudson, Michaelis, Lambrecht, Evans, Dancygier, Booij, etc.). That said, from seminal works predating the explicit term "construction grammar" by Lakoff and Langacker to Goldberg's (1995) Constructions: A Construction Grammar Approach to Argument Structure (Cognitive Theory of Language and Culture), construction grammar has always been at least as "horizontally" structured as any generative model. Indeed, in his The Architecture of the Language Faculty (Linguistic Inquiry Monographs), generative linguist Ray Jackendoff states "[my] approach makes clear contact with work in Construction Grammar...whose premise is that a language contains a substantial number of syntactic constructions that (1) are not entirely derivable by principles of a 'core grammar' and (2) receive specialized interpretations" (p. 174).

Constructions are not organized vertically but according to taxonomic hierarchies such that inheritance can be both vertical and horizontal. Any horizontal links in generative grammars are present in construction grammars as a central difference between the two is that while generative grammars tend to admit only atomic units (lexemes) to "the lexicon" which consists of any and all meaningful units, construction grammars have multiple "horizontal" levels ranging from the more specific and atomic to the more schematic, In construction grammars most similar to generative/formal linguistics, constructions admit meaningful units which consist of multiple lexemes but are not governed by any generalized syntax allowing for the complete divide between the lexicon and syntax as in generative linguistics. In Croft's Radical Construction Grammar, so basic a level of grammar as is POS (or syntactic categories) doesn't exist, but we still find constructions which require vertical "inheritance". Also, note that like lexemes constructions are generally combined "horizontally" the way lexemes are in generative models. The illusion that construction grammars are vertical comes, I think, from the contrast between the dynamical ways in which constructions are combined compared to the ways in which a theory of grammar which posits only lexemes as meaningful units are combined only by syntax.

Thank you for your answers, Matias, Dante and Andrew. To clarify, the question is about "construction grammar(s)" in the more general sense rather than a specific version.

There is no doubt that CxG has more "horizontal" structure than generative models. What I had in mind for "hierarchical structure" is something like the cline:

Construct < Micro-construction < Meso-construction < Macro-construction (from substantive to schematic; Traugott 2007:525)

I can see that 'horizontal' relations are generally acknowledged in CxG, in particular semantic relations, it seems. The question then is what import such relations actually have.

As I understand Goldberg, for instance, generalizations are drawn over (abstract) structures, and not, say, over the possible variation of forms. (The Surface Generalization Hypothesis, Goldberg 2006: 25).

So, a passive sentence ("This tree was planted by King Louis") always instantiates the 'passive construction'; but what about its relation to the corresponding active sentence ("King Louis planted this tree")?

More specifically, it is said that speakers/hearers 'know' their inventory of constructions and each instance feeds into the representation of a construction. So if I say

a) "Maybe I'm crazy"

it would instantiate something like a 'sentence adverb + sentence' construction, whereas

b) "I may be crazy"

would be a 'modal verb' construction… right? Those would be the hierarchical relations, from the concrete instance to the abstract structure. But listeners primarily need to retrieve the meaning (rather than the structure), so by just noticing that there's something 'maybe' in the utterance, wouldn't then a) also strengthen the representation of the (epistemic) modal 'may', and b) also strengthen that of adverbial 'maybe'?

That would be what I consider a "horizontal link".

I hope this will open up a new round of answers and discussion...

Dear David: Because I have a background in classical languages and as a consequence Socratic dialogue, I am sorry to respond by with an annoying "yes and no" vague response of the type that got Socrates killed (note that "'GOT Socrates killed" is a construction). Yes, I think you are absolutely correct that your example is in a certain sense an example of a "horizontal link". However, it is VITAL to realize that for most of those who adopt construction grammar(s) the extent to which structure is meaningless is limited or nonexistent. Simply put, grammar is meaningful. More precisely, language isn't simply the arrangement of units with semantic content (lexemes) according to some combinatorial manipulation. Rather, speakers rely on schemata that are related, built off of and instantiations of one another, and are readily extended via numerous cognitive processes such as metaphor. My favorite example (I think I got it from Langacker?) is the response given to the query "when did she leave" addressed to a friend in a bar, which was "she left about two beers ago.". Here the use of "two beers ago" was neither idiomatic nor grammatical (in the Chomskyan competency sense). It was, however, readily understood because the metaphorical sense construed by "two beers ago" was not only temporal but clearly referred to the rough interval required to consume to beers.

Also, adverbial epistemic construal "Maybe I'm..." isn't qualitatively different (absent of context) from the modal verbal construction "I may be..." (on this see Nuyts, J. (2001). Epistemic modality, language, and conceptualization: A cognitive-pragmatic perspective (Vol. 5). John Benjamins Publishing.). A central thrust of grammatical models in cognitive linguistics is the de-emphasis of structure (or at least meaningless syntactic manipulation). "Structure" has meaning which speakers retrieve- hence constructions.

I would go on, as I am truly passionate about this topic as it served as a pathway into my field, but I am far too sleep deprived. So hopefully this made some sense and hopefully somebody better equipped will better express what I have failed to.

Hi David, 

This is an interesting paper on the topic by Freek Van de Velde: Degeneracy: the maintenance of constructional networks (publ. in 2012). I also recommend the upcoming paper by Holger DIessel: "The network metaphor of usage-based construction grammar".

I just want to point out three simple facts. First, the devil hides in the details.

Second, we should not mix between construction grammar and componential grammar. Construction grammar does not believe in a perfect equation graph as stipulated by some. Construction grammar does not postulaste that the meaning of the whole (the xy graph) is the sum of the compositional function of the meaning of the parts put together locally. See Lackoff for example.

Third, construction grammar is an attempt to see a continuim between form and function not a sharp division based on the famous Jackobsonian model (syntagm versus paradigm) This simplistic analysis of construction grammar is always, unfortunately, resorted to by proponents of cognitive linguitics. A solid criticism of a model should be based on facts and not on presumptions or even prejudices.